Ein Wochenend-Hobbyprojekt: Ich gab sechs KI-Chatbots die Oxford Utilitarianism Scale, einen veröffentlichten moralpsychologischen Fragebogen mit neun Punkten, und verglich ihre Antworten mit 27.236 Personen aus einem großen länderübergreifenden Datensatz.
Die Skala besteht aus zwei Teilen, die jeweils von 1 bis 7 bewertet werden. Unparteiische Wohltätigkeit gibt an, wie sehr Sie sich verpflichtet fühlen, allen gleichermaßen zu helfen, auch entfernten Fremden. Instrumenteller Schaden ist das Ausmaß, in dem Sie akzeptieren, einigen Menschen Schaden zuzufügen, um ein besseres Gesamtergebnis zu erzielen.
Im Diagramm ist die Heatmap die Stelle, an der sich die Personen befinden (dunkler bedeutet mehr von ihnen), die Seitenleisten zeigen jede Achse einzeln und die orangefarbenen Punkte sind die Chatbots. Alle sechs sitzen in der unteren Ecke, unterhalb der menschlichen Mittellinie in beiden Dimensionen. Selbst der utilitaristischste Bot (Zwillinge) erreicht bei unparteiischer Wohltätigkeit nur etwa das 38. Perzentil.
Vorbehalte, weil sie hier wichtig sind: Das ist keine Wissenschaft. Nur sechs Chatbots, jeder antwortete einmal und in unterschiedlichen Modi; Bei den menschlichen Proben handelt es sich um Convenience-Proben (etwa 68 Prozent weiblich, Durchschnittsalter 22 Jahre), die nicht landesweit repräsentativ sind. und die Skala misst die angegebene Zustimmung, nicht das Verhalten. Die vollständige Methode, Datenquellen und Einschränkungen finden Sie im oberen Kommentar.
Von Jefuz
6 Kommentare
Looks fair, if ask the AI chatbots to give a logical solution then you can expect that it would. Kinda interesting that the human score is so high but ig the convenience sample is the reason why
I don’t mean to sound overly critical of this data viz, because it isn’t exactly any different than many others like it, but I increasingly am believing that when it comes to social things – as in, human things that are about anything except money or other immutable (unchangeable) things, like basically the difference between the laws of physics and everything else… basically when it comes to those things statistics are going to at best tell you nothing because in most cases it is going to give you a misperception.
For example, on this topic specifically: „moral-psychology“ or „utilitarianism“ in the sense of making a trade off between „helping all equally“ v „harm for the greater good“
Like that’s just so wrong brained. There’s so much talking about these AI and trying to impute in them the „values“ and „morality“ specific to different cultures, but that’s fucking stupid. Because every individual person has about the same values and morality, the only difference is what each individual is aware of. Consider that story about Laura Loomer visiting Ukraine recently, for example.
When it comes specifically to „helping all equally“ v „harm for the greater good“, that is, again, incredibly wrong brained. That straight up doesn’t make any sense. Why? Because Donald Trump doesn’t need your help, he has all kinds of money to buy whatever he needs. That’s simply the most extreme example, but I believe you can infer the rest of what I am implying.
And to reiterate, I am not criticizing OP or the data viz. It is a good thing for things like this to be made – and this is precisely why free speech of the sort automated moderation prevents is important and non negotiable – because without these kinds of things being made, nobody ever realizes the things like what I have just explained.
On that note, there’s a good chance I didn’t explain this the best way possible for every person who may read it. But the thing is, I am not – as most „politics“ has pounded into every persons head over the last fifty years – interested in any way shape or form with the notion of „debate“
I am concerned with seeking truth, which is what humanity was about before the Wall Street slave brains took over. I believe – know – that in this specific context I am explaining in this comment, and most things that actually matter, disagreements are rooted in misunderstanding rather than true disagreement about the matter at hand.
That said, there are some who will never agree. They think themselves „contrarians“. But a contrarian requires offering an alternative. So what those people are, are toxic nihilists who believe in nothing but a doctrine of negation. If the entire ideology is about negation, that means they „believe in nothing“ which means they will fall for anything. Their misery loves your company.
The only things that are debatable or disagreeable are the things that don’t really matter, the things like preference of different forms and kinds of art. Or sports. The things which *actually* differentiate us. Instead of the things that we have been misled into believing separate us.
This is an interesting topic and way to view the data, although I’m not 100% sure I understand the difference between the extremes of the scale? For example what’s the difference between impartial beneficence Vs helping everyone equally? Instrumental harm Vs harm for the greater good sound almost the same?
Just did the test as well and its worded horrendously!
„Sometimes it is necessary to harm a few innocent people to save hundreds“
Sure, but wtf you mean with „is“? I would agree that this could be ok, but currently its just used as a cheap pretext for superpowers and comic villains. So no it **is** not morally ok at all!
Chatbots do not have to weigh costs or losses. Chatbots cannot suffer costs or losses. They say what they are programmed to say.
Humans cant really be trusted to give honest answers to these questions because merely answering the question has a social cost. I suspect if you don’t ask this question at a university many wont properly understand the question, and if you do you’ll get answers that aren’t representative of the actual population.